Predicting utilization of emergency contraceptives in ethiopia and identifying its predictors using machine learning

Abstract

Abstract Despite policy support, inappropriate use of emergency contraception in Ethiopia contributes to high rates of unintended pregnancy and maternal mortality. Traditional statistical analyses have struggled to identify complex predictors. This study used machine learning and Explainable AI to improve the prediction and interpretability of emergency contraception use. We analyzed data from 2,334 women in the PMA Ethiopia 2023 survey. Eight ML algorithms were tested to predict past-year emergency contraception use (4.4% prevalence), and the SMOTE was used to address class imbalance and SHAP values for interpretation. Logistic Regression on SMOTE data achieved the best performance (AUC-ROC: 0.85; Recall: 0.85; precision:0.72). The most important predictor was emergency contraception awareness (“heard_emergency”), followed by media exposure and family planning discussions at health facilities. Conversely, recent reproductive events such as unintended pregnancy were linked to non-use. Static demographic factors showed poor predictive value. Findings highlight that knowledge gaps, rather than poverty or physical access, are the key barriers to emergency contraception use. Tailored media campaigns and routine health counseling could enhance emergency contraception uptake. ML and XAI offer powerful tools for guiding targeted reproductive health interventions. Knowledge of emergency contraception is the strongest modifiable predictor of use, ranking above socioeconomic and geographical factors. Media exposure (radio/TV) and quality health system contact were found to be key complementary drivers. Methodologically, extreme class imbalance was tackled using the synthetic minority oversampling technique-enhanced Logistic Regression to achieve robust predictive performance (85% recall), and shapley additive explanations analysis uncovered actionable intervention levers.